fix(checkout-api): resolve checkout-race-condition

[Spkap]

Summary

Resolve checkout-race-condition for checkout-api with a validated ReplayX patch candidate.

Changed Files

• demo_app/src/inventory/reserve-stock.ts
• demo_app/src/checkout/submit-order.ts

Validation

• Concurrent checkout repro: exit 0
• Sequential checkout sanity: exit 0

Rollback

Revert the live inventory guard and concurrent checkout settlement handling.

[Copilot]

Pull request overview

Fixes the demo app’s checkout-api oversell race by validating inventory availability at commit time and by treating expected OutOfStock failures as non-fatal in the concurrent repro path.

Changes:

• Add a post-delay “commit-time” inventory availability guard in reserveStock to prevent oversell under concurrency.
• Return reservation metadata (token/version) from the committed inventory record rather than the pre-check snapshot.
• Update the concurrent scenario runner to use Promise.allSettled, ignore expected OutOfStock rejections, and return only successful checkout results.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
demo_app/src/inventory/reserve-stock.ts	Adds commit-time stock validation and returns committed snapshot/version metadata.
demo_app/src/checkout/submit-order.ts	Makes concurrent repro tolerant of OutOfStock by using allSettled and filtering expected failures.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The inline comment about the delay being an “Intentional bug” is now misleading/outdated. With the new post-delay record.available guard, the stale pre-check no longer causes oversell on its own, so the comment should be updated to describe the delay as a concurrency simulation while commit-time validation prevents stale snapshot writes.

[Copilot]

This logic treats stock exhaustion as a “soft” failure by checking Error.message.startsWith("OutOfStock:"). Keying behavior off an error message prefix is brittle (message changes, localization, non-Error throws) and can misclassify failures. Prefer throwing a dedicated OutOfStockError (or an Error with a stable code property) from reserveStock and checking that here.

[Copilot]

With this change, the concurrent scenario no longer throws when one worker hits OutOfStock, so demo_app/scripts/repro-checkout-race.ts will exit 0 for the concurrent repro. The incident fixture incidents/checkout-race-condition.json still marks that command as the failing repro with expectedExitCode: 1, and the orchestrator uses that expectation to determine whether the repro is confirmed. If the intent is to fix the incident in-place, update the incident command expectations (or swap which command is considered failing vs healthy) to keep the pipeline/demo consistent.

Suggested change

	const hardFailure = settled.find(
	(result) =>
	result.status === "rejected" &&
	!(result.reason instanceof Error && result.reason.message.startsWith("OutOfStock:"))
	);
	if (hardFailure && hardFailure.status === "rejected") {
	throw hardFailure.reason;
	const rejectedResult = settled.find((result) => result.status === "rejected");
	if (rejectedResult && rejectedResult.status === "rejected") {
	throw rejectedResult.reason;